To our knowledge, this is the most extensive examination of the association between blood culture testing and bloodstream infections on hospital mortality. We found that blood culture testing was common and that bloodstream infections were detected in almost 2% of hospitalizations. Patients undergoing blood culture testing were much sicker and had a notably higher risk of death in hospital that peaked around the time of their test. Even after adjusting for important confounders, patients with negative blood cultures still had a significantly increased risk of death in hospital. The risk of death was higher still in those with bloodstream infections with the risk being highest in bloodstream infections that: were detected in the first hospital day; were polymicrobial; occurred during a neutropenic episode or while exposed to immunosuppressants; or those due to Clostridial and Candidal organisms.
Our study has several important findings. First, we found that the independent risk of death in hospital increased significantly whenever blood cultures were ordered, even when those blood cultures did not grow a microorganism. This indicates that patients undergoing blood culture testing are sicker than others, even after adjusting for measured covariates that distinguished these patient groups (Table
1). Obviously, we do not believe that the actual act of measuring blood cultures increases the risk of death in hospital. Instead, we suspect that an increased risk of subsequent bad outcomes regardless of the blood cultures result is due to the test indicating a sicker population than is indicated by the measured covariates. We believe that this phenomenon is likely true for other tests (such as electrocardiogram, portable chest radiograph, cardiac enzymes, and arterial blood gases) that are done when patients deteriorate acutely. This observation should be considered when analyzing the influence of abnormal test results on hospital outcomes. Second, we found that the increased independent risk of death in hospital associated with negative blood cultures and bloodstream infections was maximal at the time that the blood culture was procured and decreased significantly over time. This likely reflects the benefit of treatments given to surviving patients. Finally, the harmful effect of bloodstream infections was highest with particular microorganisms (notably Clostridial and Candidal) and in immunocompromised hosts (those exposed to immunosuppressive agents and those with neutropenia). These results confirm how such patients with bloodstream infection must be treated aggressively.
Our study has several interesting comparisons with previous analyses of bloodstream infections and hospital mortality. The distribution of microorganisms that we identified in our cohort was very similar to that identified in previous analyses
[2, 5, 12, 13, 15]. Similar to our results, several other analyses have found particularly high mortality rates in patients with Candidal bloodstream infection
[5, 6, 14–16]. To our knowledge, ours is the most extensive analysis that included patients without blood cultures and those with negative blood cultures; this characteristic is necessary to precisely gauge the influence of bloodstream infection on mortality risk relative to other patients and independent of confounders associated with the actual measurement of blood cultures. Finally, several studies had previously found – in contrast to our study – that mortality risk was higher in those with nosocomial bloodstream infections
[2, 12, 14, 16]. We believe that these analyses are susceptible to time-dependent bias
 since patients must remain alive in hospital for a specified period of time to be classified with nosocomial bloodstream infection. Since patients who remain in hospital longer tend to be sicker, such analyses could falsely attribute mortality risk from confounders in these patients to nosocomial bloodstream infection. Analyzing bloodstream infection as a time-dependent covariate within a survival model – as we did in this analysis – avoids this potential bias
Our study has several notable attributes. We captured all hospitalizations, all blood cultures, and all bloodstream infections at our hospital during the study period. Our statistical model recognized the time-dependent nature of blood culture testing, bloodstream infections, and their association with death in hospital. However, several potential limitations of our study should be kept in mind. First, although we found that bloodstream infection was associated with an increased risk of death in hospital, we have no way of determining whether the infection actually caused the death. Primary data review would be needed to determine – if possible – whether or not the bloodstream infection caused a particular patient’s death. Such analyses are necessary to determine if and how we might intervene to improve outcomes with hospital-associated bloodstream infections. Second, our data did not account for treatment of bloodstream infections. Outcomes in patients admitted for community acquired pneumonia are improved significantly in those who receive antibiotics more quickly
. Berjohn et al.
 found that patients with pneumococcal bacteremia receiving least 1 active antibiotic within 4 hours of blood cultures were significantly less likely to die in hospital (odds ratio [OR], 0.47; 95% confidence interval [CI], 0.2-1.0). It is possible that timeliness of appropriate antibiotics - and other interventions to control infection - could have an independent influence on death risk in all patients with bloodstream infections. Third, we did not have access to information about potential sources or causes of bloodstream infections, such as catheters. It is possible that mortality risk associated with bloodstream infections may change significantly based on the presence of foreign bodies. Fourth, the utilization of blood cultures requires a physician’s response to clinical data input and are not – by themselves – a pathophysiological marker. Physicians will vary in their response to various clinical data and, as such, will have different thresholds or likelihoods for ordering blood cultures. Therefore, the external validity of our findings to other centres could be questioned. However, supporting the generalizability of our findings is the large size and long duration of the study as well as its complete capture of all blood cultures at our large institution, all of which will ensure a large number of different physicians who were captured by the analysis.